Run 01a016ae · KernelIndex

Primary measurement

621.5µsmedian

Rank 4 in its comparison group · source-native comparison · observed 2024-09-03

Compare with #1 →

ms 20 0.6208000183105469 · ms 50 0.621504008769989 · ms 80 0.6223679780960083

Identity

implementationTransformers

projectHugging Face Transformers

revisionunknown

workloadseq = 2048 · hidden = 8192 · q_heads = 32 · kv_heads = 8 · bf16

comparison keysha256:63b583a3afe1e4d7…

sourceLiger-Kernel benchmarks

external idrope/backward/huggingface/nvidia-a100-sxm4-80gb/hidden8192-kv_heads8-q_heads32-seq2048-bf16/2024-09-03-15-39-59

sha256:c3252a98ba487075f12047e…

Correctness

Marked passed by the source; the correctness policy was not published.

Workload

seq2048

hidden8192

q_heads32

kv_heads8

kbf16 [1, 8, 2048, 256]

qbf16 [1, 32, 2048, 256]

definition comparatornot_asserted

Measurements

latency · median621.5 µs

latency · p20620.8 µs

latency · p80622.4 µs

Protocol

harnessLiger-Kernel benchmark scripts

timercuda_events

primaryStatisticmedian

comparabilityFamilyliger_kernel_bench

Environment

gpuNVIDIA A100 (sm_80)

Artifacts

No artifacts published with this run.

Replications and notes

No attestations yet.

Community attestations. They never change the evidence level; only a KernelIndex-controlled rerun does.

Add a reproduction or note

Canonical manifest

Show manifest

{
  "run": {
    "kind": "BenchmarkRun",
    "spec": {
      "status": "passed",
      "timing": {
        "latencyNs": {
          "median": 621504
        },
        "primaryStatistic": "median"
      },
      "observedAt": "2024-09-03T15:39:59.000Z",
      "measurements": [
        {
          "unit": "ns",
          "value": 620800,
          "metric": "latency",
          "statistic": "p20"
        },
        {
          "unit": "ns",
          "value": 622368,
          "metric": "latency",
          "statistic": "p80"
        }
      ],
      "sourceNative": {
        "source": "liger-kernel-bench",
        "metrics": {
          "ms_20": 0.6208000183105469,
          "ms_50": 0.621504008769989,
          "ms_80": 0.6223679780960083
        },
        "benchmark": "rope/backward",
        "externalId": "rope/backward/huggingface/nvidia-a100-sxm4-80gb/hidden8192-kv_heads8-q_heads32-seq2048-bf16/2024-09-03-15-39-59"
      },
      "protocolDigest": "sha256:080e6e66e4ec75a131aa0c9121c74b8e8eed64e7b6d621d567c6a28f79548be8",
      "workloadDigest": "sha256:c66457140cb6cfc6fdcb24a3187286a041f2b04f37522e1bd65089aacf4cd0f3",
      "environmentDigest": "sha256:90c9e2764222f592477cfccf3373519add5a1b881563956ad058c4f99e41b6b9",
      "implementationDigest": "sha256:459baf6556e5536285c26b26d11eda4980b44d362142461d8e7e5b4f3900ce23"
    },
    "metadata": {
      "name": "liger-rope-backward-huggingface-nvidia-a100-sxm4-80gb-hidden8192-kv-heads8-q-heads32-seq2048-bf16-20",
      "title": "Rotary position embedding · huggingface · backward",
      "labels": {
        "liger_version": "0.2.1"
      }
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "protocol": {
    "kind": "BenchmarkProtocol",
    "spec": {
      "harness": {
        "name": "Liger-Kernel benchmark scripts",
        "repository": "https://github.com/linkedin/Liger-Kernel"
      },
      "measurement": {
        "timer": "cuda_events",
        "primaryStatistic": "median"
      },
      "comparability": {
        "notes": "Timed: the backward pass of the module output given a random upstream gradient, via triton.testing.do_bench with quantiles 0.5/0.2/0.8 (median with p20/p80 spread). The upstream CSV records no CUDA, driver, or torch version; the Liger release rides each run's labels. Comparable only within one kernel, workload, timed pass, and GPU.",
        "family": "liger_kernel_bench"
      }
    },
    "metadata": {
      "name": "liger-bench-do-bench-backward",
      "title": "Liger benchmark harness · backward pass"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  },
  "environment": {
    "kind": "ExecutionEnvironment",
    "spec": {
      "hardware": {
        "vendor": "nvidia",
        "product": "NVIDIA A100",
        "formFactor": "SXM4-80GB",
        "architecture": "sm_80"
      },
      "software": {}
    },
    "metadata": {
      "name": "liger-bench-nvidia-a100-sxm4-80gb",
      "title": "Liger benchmark host · A100-SXM4-80GB"
    },
    "apiVersion": "kernelindex.dev/v1alpha1"
  }
}

Cite this record (permalink, digest, access date)

Report an issue with this run

Published 2026-08-18 · Liger-Kernel benchmarks · BSD-2-ClauseAll results for Rotary position embedding →JSON